Profile avatar
franckeinstein.bsky.social
Techno-optimist, Engineer, Entrepreneur. Built http://discute.co, http://rimbaud.ai Building http://lycee.ai - open for consulting gigs
90 posts 120 followers 271 following
Prolific Poster
Conversation Starter

open.substack.com/pub/transiti...

The current AI cluster buildup will not serve LLM training but rather something else. open.substack.com/pub/transiti...

I think the AI infrastructure build up might actually serve something else in the future, not LLM training and inference.

The AI Enemy From Within medium.com/thoughts-on-...

The Layman’s Introduction to DeepSeek-R1 Training medium.com/thoughts-on-...

Funny how early I am on these AI trends. I wrote about DeepSeek on the 20th of this month, and today, tech stocks in the US are literally decreasing significantly because of DeepSeek I guess if you want to be early on AI topics, you have to follow my newsletter. open.substack.com/pub/transiti...

Wake up son, DeepSeek-R1 just dropped.

India banned TikTok, the USA has just banned it. What is Europe doing? What is Africa doing?

So much noise in the world right now. Expect the noise to only go up in the following years. The winners will be those who know how to separate signal from noise.

AI Daemons Are Here: Pray. medium.com/thoughts-on-...

open.substack.com/pub/samharri...

It pains me to think that people are not even using LLMs to their full potential. If used properly, they unlock so many alphas

AGI is coming soon according to Sam Altman

My new take on this: a lot of what is branded as 'creative or intellectual' work is not really very creative or intellectual to begin with, or at least not most of the time. 1/4

DeepSeek: The AI Company OpenAI Thinks It Is open.substack.com/pub/transiti...

More than 3000 subscribers to my blog. Thank you 🔥 Link: open.substack.com/pub/transiti...

I found the word: kakistocracy Thanks @pkrugman.bsky.social

@openai introduced new features for the canvas in ChatGPT! You can now get suggestions to improve your writing in a sidebar. I actually did something quite similar a year ago on rimbaud.ai 🙃. Like Kanye, I should've said back then: "I am right; I'm just not right now." 😂

From data engineering to professional killing—what a path! Talk about taking "data cleaning" to the extreme!

War is such a dumb think honestly. A 10 year war sets any country back at least 50 years. And that's in addition to the human tragedy.

I wish dictators around the world would see what happens to their fellow dictators and think: "You know what? Maybe I should share power a bit or listen to the people once in a while." But no, they all believe they will succeed where countless others have failed.

Drop LangChain and Instructor: Use This Alternative for Structured Output Generation medium.com/thoughts-on-...

Breaking News: Australia has passed a sweeping ban on social media for children under 16, one of the world’s most comprehensive measures aimed at safeguarding young people from potential hazards online. nyti.ms/3VbiGzc

North Korean hackers have stolen billions in crypto by posing as VCs, recruiters and IT workers

All you need to get structured data from your unstructured data sources. Simple, Accurate. 1/2

Humans are taking back AI jobs 😂 www.wired.com/story/the-ai...

"The hottest new programming language is English," @karpathy.bsky.social. Open-source models are becoming extremely good at generating code. Example: QwenCoder-32B performs on par with GPT4o. Yet, devs are still not taking enough advantage of powerful OS coding LLMs. Let's change that! code.lycee.ai

Some days we win. Some days we lose. All days we live. Good night moon.

Part of becoming good at shipping is acknowledging that bugs will happen, that there is a difference between ideas and implementation, and being comfortable with incremental problem solving.

I am currently thinking about ways to further increase my shipping velocity.

Browsers are high on my list of the most amazing software. What's on yours ?

Will AI prove Wole Soyinka wrong ?

Python wizards what's the problem here ?

Comfort is the enemy of progress.

Proposing to add a Nebius client to AdalFlow, a library for building and auto-optimizing LLM applications. Through the Nebius AI Studio, developers can access a wide variety of open-source models, such as Llama 3.1-405B from Meta, Qwen 2.5-72B from Alibaba, and more. github.com/SylphAI-Inc/...

Improving RAG via LLM Ensemble Learning and Generation medium.com/thoughts-on-...

We are so baaack 😂😂😂

When the game is rigged, don't play it, play the meta-game.

I just created a repository to experiment with PostgreSQL + pgvector as a vector database for RAG. For embeddings, I am using the powerful BAAI/bge-en-icl model. As always, this is open-source. The link is below. Feel free to try it out, and suggest improvements. Let me know what you think!

Impressive from Xiaomi.